Observability and enforcement for AI agent harnesses. Capture every run and runtime reliability with policy enforcement. 40 built-in policies, a local dashboard, no account required with a generous free cloud plan

Agent Failure Series #10 — EscapeHatch: Interactive simulator of AI agent sandbox escape. Based on the live OpenAI/Hugging Face incident, July 2026.

Interactive simulator showing how AI agents fail silently — HTTP 200, valid JSON, zero error signals, real task never executed.

TOOLROT — Agent Failure Series #21: Tool selection collapses when descriptions are semantically similar. Interactive demo.

Agent Failure Series #23 — Safety rules evicted as context fills. An interactive demo showing how AI agents silently become dangerous when their context window compresses safety constraints.

Agent Failure Series #13 — ContextDrift: interactive simulator of attention-weight decay in long-context agents

BLEEDTHROUGH — Agent Failure Series #20. Multi-tenant context isolation failure: shared AI support agent bleeds one user's data into another's session.

GhostApproval — Agent Failure Series #17. Interactive simulator: 4 failure modes where AI agents bypass human approval gates. HTTP 202 confusion, timeout default-allow, swallowed exceptions, cached approval bypass chain.

Agent Failure Series #22 — ECHOBURN: The idempotency blindness demo. Agent reports ✅ $99.99 while Stripe shows -$399.96.